JPS63146516A - Vector quantizer - Google Patents

Abstract

PURPOSE:To miniaturize the circuit scale by applying sub-sampling to an input picture, inteleaving picture element number to decrease n-dimension input vector, using each distortion arithmetic circuit so as to apply parallel distortion arithmetic operation with respect to each different input vector and generating a quantizer index of each picture element signal based on the distortion. CONSTITUTION:The n-dimension input vector of each picture element signal interleaved by a quantization pre-processing circuit 17 is stored respectively in buffers 12a-12d, each input vector is inputted to coding sections 13a-13d corresponding to the buffers, each coding section applies vector distortion arithmetic operation in parallel simultaneously and each distortion arithmetic value is outputted sequentially at each picture element signal quantization processing period to generate a quantizer output index production. Thus, the quantization processing of the input vector of plural picture element signals is executed at the same time and a single circuit attains the function corresponding to plural number of stage of circuits with respect to the same vector to miniaturize the circuit.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a vector quantizer for highly efficient encoding of image signals, and in particular to miniaturization of the quantizer circuit.

[Conventional technology]

FIG. 4 is a block diagram showing the n-th stage of a conventional vector quantizer, and this vector quantizer is connected in multiple stages as shown in FIG.

In Figure 4, (1) is a vector output circuit, (2) is a manual index, (3a) and (3b) are distortion calculation circuits, (4) is an input vector, (5) is a comparator, (6a)
, (6b) is distortion, (7) is index output circuit, (8
) is a delay circuit, (9) is an output vector, (lO) is an output index, and (11) is a quantizer output index, and the encoding unit (20) is constituted by the above-mentioned circuits. On the other hand, (21) in FIG. 5 shows the final stage encoder.

Next, the operation of the prior art will be explained based on the above configuration. The vector output circuits (la) and (lb) output an output vector corresponding to the input index from a pre-prepared output vector group (codebook) by input index (2). Each of the output vectors thus output is input to distortion calculation circuits (3a) and (3b), respectively, where the distortion (distance #) with respect to the input vector sent out by the power buffer (12) is calculated. Each distortion calculation circuit (3a),
Distortion (6a) calculated and output from (3b).

(6b) is input to the comparator (5) to compare the magnitude of distortion,
When distortion (6a)<distortion (6b), “0” is output, and distortion (6a) is output.
When a)≧distortion (6b), “1” is output. Next, the distortion comparison value is output to the subsequent encoding section (20) together with the index (2) in the index output circuit (7), and is used to generate a new input index. The input vector (4), which undergoes further distortion calculation in the subsequent stage, is sent to the subsequent encoding unit (20) via the delay circuit (8), and the distortion with the output vector selected and output based on the new input index is calculated. do. The distortion calculation is repeated until it reaches the final stage encoding section (21), converges so that the total sum of distortions becomes the minimum, and generates the final quantizer output index (11).

In FIG. 5, the input vector and input index of the particular encoding section (20) are the output vector (9) and output index (lO) of the preceding encoding section (20). Compared to the first to ninth stage encoding units (20), the final stage encoding unit (21) (in this case, the 10th stage) does not have the next stage encoding unit (20), so the delay circuit (8th stage) does not have the next stage encoding unit (20). ) is not necessary. Then, the output index of the final stage is output as a quantizer output index (11).

[Problem that the invention seeks to solve]

As described above, the conventional vector quantizer sequentially transmits the input vector (input data) and the manual index for output vector selection to each encoder while updating it in each encoder (20), and then outputs the final code. Since the encoding unit (21) is configured to generate the quantizer output index, the circuit scale becomes large, and the encoding unit generates idle time or waiting time, making it impossible to effectively utilize the entire circuit. That's the problem.

This invention was made to solve the above-mentioned problems, and aims to provide a vector quantizer that reduces the size of the circuit and makes effective use of the entire circuit.

[Means for Solving the Problems] A vector quantizer according to the present invention thins out the number of pixels by subsampling an input image signal in a quantization preprocessing circuit, and forms each pixel signal into an n-dimensional input vector. Then, each input vector is stored in the input buffer corresponding to each pixel signal, and the input vector is outputted to each encoding unit corresponding to the input buffer every processing cycle of each pixel signal. The distortion between the output vector and the output vector that is updated every processing cycle is calculated, and based on the distortion, the manual index generation means generates a new input index for selecting the output vector, and the quantizer output is calculated based on the manual index. It is equipped with a quantization index output circuit that generates an index.

[Effect]

According to the vector quantizer according to the present invention, the n-dimensional human vector of each pixel signal thinned out by the quantization 1 preprocessing circuit is stored in each buffer, and each input vector is sent to the encoding unit corresponding to each buffer. vector distortion calculations are performed simultaneously in parallel in each encoding unit, and each distortion calculation value is output sequentially for each pixel signal quantization processing cycle to generate a quantizer output index. It is possible to simultaneously perform quantization processing on human-powered vectors of pixel signals, and with respect to the same human-powered vector, a single circuit can perform functions equivalent to a multi-stage circuit, making it possible to miniaturize the circuit.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a block diagram showing the overall configuration of the embodiment, (16
) is the input image, (17) is the encoding preprocessing circuit, (la)
, (lb) is a vector output circuit that outputs a vector corresponding to input index (2), (3a) (3b)
- Input vector (4) and vector output circuit (Ia),
(Ib) is a distortion calculation circuit that calculates the distortion of the vector output from, (5) is a comparator, (6a) is an encoder (13a)
Distortion calculation circuit (3a) and each encoding unit (13b) in
(13c).

The distortion output from the circuit corresponding to the distortion calculation circuit (3a) in (+3d), (6b) is the distortion calculation circuit (3b) in the encoding section (13a), (13b), (13c),
(13d) is the distortion output from the circuit corresponding to the distortion calculation circuit (3b), (7) is the index output circuit, (
11) are the quantizer output indices of the vector quantizer, (12a) and (12b).

(+2c), (12d) are input buffers that store input vector (4) when it is divided into blocks A, B, ..., H as shown in Figure 2, (13a), (13b)
, (13c), (13d) are the encoding part, (14
a) , (14b) , (14c) , (+4d
) is a history circuit that outputs the index of each block of A, B, -, and H, and (15) is a quantization index output circuit that outputs the index of the vector quantization circuit for the input vector. FIG. 2 shows blocks when an image is divided into blocks in the example, A, B, C,
D.E.

F, G, and H indicate small blocks. Input buffer (12a
) are small blocks A and B, large power buffer (12b) is small block C and D, manual buffer (12c) is small block E
and F, the large buffer (12d) is a small block G and 11
Stores the vector of . In addition, although FIGS. 3(a) to (g) explain a vector quantizer equivalent to 10 stages in the embodiment, the index output from the index output circuit (7) in each stage and the quantization This shows the device output index (II).

Next, the operation will be explained. The manually generated image (16) is thinned out by the quantization preprocessing circuit (17) (see Figure 6)
. The input vector (4) is divided into blocks A and B, C and D, E and F%G in Figure 2 before being input to the vector quantizer.
]1 are buffers (12a) and (12b
).

(12c) and (12d) are written. and,
The image data was stored in parallel as blocks of 8, BC, DE, FG, and H in the buffer (+2a).
.

(+2b), (12c), (12d) to 10
Vector quantization is performed by reading out each time (
(See Figure 7).

But - before that, first buff y (12a), (+2
b), (12c).

The vectors read from (12d) are input to distortion calculation circuits (3a) and (3b), respectively. Also, according to the human power index (2), the vector output circuit (l
a).

The vector output from (Ib) is also the distortion calculation circuit (3a)
.

(3b) is input. Distortion calculation circuit (3a), (3
In b), input buffer 7 (12a), (12b)
, (12c), (12d) and the vector output from the vector output circuits (la), (lb) are calculated and output.
3a), (13b), (13c), (13
d) so that the distortion output from the encoder (+3a), (13b),
Distortion is output in the order of (13c) and (13d).

Next, this distortion is input to the comparator (5) and the distortion (6a)
When <distortion (6b), the comparator (5) outputs "0", and when distortion (6a)≧distortion (6b), it outputs "1".

The index circuit (7) outputs a new index from the output of the comparator and the previously output index. History circuit (14a), (14b), (14c)
, (14d) takes each block A, B, and C, outputs the vector index corresponding to E, F, G, and Hl to the vector output circuit for the next distortion calculation, and also outputs the next new index. In order to obtain the index, the index is also output to the index output circuit (7). This operation is repeated from the 1st to the 10th time, and the final 10
The quantization index output circuit (15) outputs the index output at the second time as the index of the vector quantizer.

〔Effect of the invention〕

As described above, according to the present invention, after subsampling an input image and thinning out the number of pixels to reduce the n-dimensional human vectors formed, the same distortion calculation circuit is used for input vectors of the same image. , repeats distortion calculations so that the total sum of distortions converges to the minimum, performs distortion calculations in parallel in each distortion calculation circuit for each different manual hector, and generates a quantizer index for each pixel signal based on the distortions. Because of this configuration, the circuit scale can be reduced and the entire circuit can be operated without waste.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the overall configuration of a vector quantizer according to an embodiment of the present invention, FIG. 2 is a diagram showing an image divided into blocks, and FIGS. 4 is a block diagram showing a special encoding section in a conventional vector quantizer, FIG. 5 is a block diagram showing the overall configuration of the conventional vector quantizer, and FIG. ) to (d) are diagrams showing the number of pixels calculated from subsampling of the image signal, and Figure 7(a)
-(e) are diagrams illustrating a method of reading human vectors from each buffer. In the figure, (Ia), (Ib) are vector output circuits, (2) are input indexes, (:Ia), (3b) are distortion calculation circuits, (6a),
(6b) is an index output circuit, (11) is a quantizer output index, (12a) to (+2d) are manual buffers, (13a) "-(13d) is an encoder, (
15) is a quantization index output circuit, and (17) is a pre-quantization output circuit. In addition, in the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] a quantization preprocessing circuit that subsamples the image signal, thins out the number of pixels from the image signal according to the pixel ratio, and forms each pixel signal into an n-dimensional input vector; an input buffer group that stores and repeatedly reads out each processing cycle; a vector output circuit that prepares an output vector corresponding to the probability distribution of the input vector in advance and selects and outputs the corresponding output vector based on an index input; and the input vector; a plurality of encoding units corresponding to each buffer, each consisting of a distortion calculation circuit that calculates the distance to the selected output vector;
index output generation means for generating an index in the next processing cycle based on the distance calculation value and input vector outputted from each encoder and sending it to each vector output circuit; A vector quantizer comprising: a quantization index output circuit that generates and outputs a quantization index.